Explore ways to reduce code size of match bodies

[morganthomas]

As a leanVM developer, I want to reduce the compiled code size of match bodies, so that the program trace takes less time to commit to. There are ways that we could reduce the compiled code size of match bodies. In particular:

1. When arms are very different in size, extract each arm into a separate label, and make the match table just consist of jumps to those labels.
2. When useful, place other code blocks into the unreachable portions of arms which are currently filled with padding.

[morganthomas]

I am working on this. My plan is to try to place code blocks into the padding of arms.

[morganthomas]

I think a good approach is to use a greedy algorithm, which tries to place the largest possible block in the smallest gap it will fit in, replacing each gap used with the unused remainder (if any) of the gap. However, I am noting that this does not make sense to prioritize currently, as the code size does not appear to be a major factor in proving time, and in any case, code size is rounded up to a power of two, but our current code sizes are not near enough above a power of two that this optimization would make a difference in my estimation.